You need to provide some examples of the kinds of images that didn’t match, and preferably post some of them for examination. Were they images containing text? If so, were you using the Text search type? Were all the machines using the same Font Smoothing Setting (Automatic doesn’t count)? Were the machines that you originally captured the images on running the same version of the OS? That doesn’t matter for most elements, but there are some exceptions.
Of the things that you mentioned setting the same, the only one that might really matter is the desktop background, and then only for those elements that sit directly on the desktop. The display/resolution doesn’t matter (except where something that used to fit on the screen doesn’t anymore) and unless one of the machines was set to greyscale, the color settings aren’t likely to cause problems either.
Unfortunately you probably captured the new images over the old images – that made sense, but the best evidence would have been to compare the old images directly to the recaptured images. It is definitely possible to make your scripts transportable across multiple machines, either by figuring out what the differences might be, or by using image collections to deal with discrepancies.